
GEO vs SEO Explained: Myths, Query Fan-Out & AI Search Reality
Let’s face it the debate about GEO vs SEO is still pretty noisy.
On the one hand there are a group of people that say GEO is entirely different and requires a completely different approach.
On the other hand, there are a group of people (including Google) that say good GEO is really just good SEO.
So which is it?
Well, there are subtle nuances here, let’s dig into it!

- The Elephant in the room -> Google
- Most LLM Searches involve regular web search to get their answers
- Page Rank and Quality – Something the LLMs Fundamentally Lack
- Generative AI and Hallucinations
- It’s not all Google
- The Numbers
- It’s the early 2010s again!
- GEO / AEO Myths
- Acronym Soup
- References & Resources
Elephant in the room -> Google
OK so the elephant in the room is Google.
Google Search has AI Overviews which are called AIOs. AIOs behave much like other SERP (Search Engine Results Page) features – such as a featured snippets, in fact AIOs have largely displaced featured snippets.
Google also has AI Mode which is a different more conversational search interface a lot like the current swathe of LLM (Large Language Models) model based tools.
Google also has Gemini which directly competes with ChatGPT, Claude and other LLM based tools.
The big difference between Google and the other providers is Google has the data and infrastructure of its core search engine beneath it. This is also something many of their competitors leverage. See what Google says on Generative AI.
Q. “Is SEO still relevant for generative AI search?”
- “In short, yes!”
In 3 words Google settled the debate.

Most LLM Searches involve regular web search to get their answers
Most AI searches actually involve many regular searches you don’t see.
It is quite the irony that in order for the LLM tools to be accurate they need to use many regular web searches to pull back information in which to ground and contextualize their answer. When they don’t it goes wonky quite quickly (more on that later).
The crux of this is something referred to as Query Fan Out. The intention of Query Fan Out is to do several web searches for topics related to the core query – (getting less relevant and less specific not more – in my view this is a move in the wrong direction and is not an advancement in search technology) – then when they have this information they bring it back and ground the original query in all of this context.

So when you do one LLM query you actually end up doing multiple Google searches to get the information.
This happens behind the scenes, but it is quite the irony that in order for these tools to work well, they need to resort to regular web search engines to get their answers.
I often say “if only we had a system of indexing and a mechanism to assess quality and rank accordingly …. Oh wait we do, it’s called a search engine” – Trevor Stolber
Page Rank and Quality – Something the LLMs Fundamentally Lack
This brings me to PageRank (PR) and quality – something the LLMs fundamentally lack.
PageRank is Google’s foundational breakout search technology that used links from websites as a voting system. They seeded some core set values and let it propagate throughout the web, assigning every link a certain PR value which counted as the strength of that vote.
Through decades of search advancement, they got better at understanding context and relevance.
This is something that the LLMs fundamentally lack.
Generative AI and Hallucinations
Lets not forget that GenAI or Generative AI, which is what the LLM tools are, is literally designed to make things up.
Lets repeat that and pause – they are designed to make things up – that is literally in their design.
Search engines retrieve; LLMs generate (make things up).
LLMs are just fancy predictive text machines – they know what the most likely thing to come next is – that is really all they are. There is some very clever use and adaptations of this but at a base level they are predictive text machines not an advanced intelligence tool that many would lead you to believe.
They most definitely do not think.
It’s not all Google
We should recognize that it’s not all Google. The battle is fierce and trends and adoption of models change frequently.
ChatGPT and Claude are major players that occupy a lot of the AI landscape
Bing has a place too – with the Microsoft ecosystem being a major player. Microsoft’s index is used in ChatGPT as well as copilot and any Bing based websearch that leverages generative AI. Don’t under estimate the importance of this on corporate networks and in industries that are highly aligned with the Microsoft Tech Stack.
The Numbers
The numbers don’t really support the hype when you break it down.
Companies are allocating roughly 27% of their marketing budgets for AI however what AI delivers is much lower than this so it’s a disproportionate spend.
The number of users and user growth are drastically higher than those actually using the AI tools for traditional search.
It’s the early 2010s Again
Those of us who have been in the SEO industry for a while will recognize a lot of the “tactics” and “techniques” spun by the GEO/AEO crowd.
SEO (to some) used to be a cat-and-mouse game where they would try and trick the search engines, it would work for a while and be abused, then an algorithm update would come along and completely decimate the sites that had been using the tricks to get rankings.
Most businesses can’t tolerate that risk to a key traffic source – so be very careful with what and who you employ to do your search marketing for you.
GEO / AEO Myths
The great irony of a lot of the GEO advice is that it tells people to do to their content what the LLMs do when they ingest content.
They strip a lot out code, structure and even words. Your text gets heavily trimmed and broken down into tokens or chunks (hence the chunking “tactic”).
Chunking
Chunking is something that the LLMs do when they parse (read) your text, so there is literally no point in doing this. Google is also very good at reading sections (passage indexing) rather than the whole content (page indexing) when a certain section of a page is relevant.
Many of the GEO guru suggestions is to do something that is already done, it is completely irrelevant and pointless.
Unfortunately, in this industry, that doesn’t stop people from writing (and charing for) extremely bad advice.
Mention important topics often and early – this works for SEO & GEO and is logical. Why would you wait until the end of an article before mentioning a key point?
Short segments
There are reasons for keeping your sentences and paragraphs relatively short. Even traditional organic algorithms struggle with longer sections of text, despite how good Natural Language Processing (NLP) has become – it’s truly amazing!
However trimming and shortening your content is not time well spent. The LLMs will do this automatically- you don’t need to. You just wasted some time that’s all.
Listicles
Listicls weere a “technique” that the GEO/AEO bros recommended for a minute, and for a minute they worked. Its actually really disappointing to see that this did work, but it also exposes how weak the base technology is when someone can throw up a self hosted completely fake list and it influences the output of an LLM.
There are some truly hilarious examples of this going very awry.
One of the funniest things that happened was that the LLM tools recognized this abuse and started to rate the second one in the list higher than the first, often ignoring and not highlighting the one that was listed first. Kind of hilarious.
These are some of the cat-and-mouse tricks I referred to. They get shut down quickly and relying on and applying these type of techniques is a recipe for disaster.
So most of the people that did this ended up shooting themselves in the foot.
LLMs.txt
Mark Williams-Cook did a very interesting experiment to debunk the myth that LLMs.txt were necessary or helped in some way. It is an interesting option but it does not really do very much, it does not do what many people claim it does, it does do what a normal text file will do (get indexed and be discoverable in search). To show this Mark made up the cats.txt standard and showed that it was cited and recommended in AI based searches. Bare in mind this was completely deliberate and shows how easy manipulation is.
Schema
Schema is a very interesting thing. Fundamentally its a structured way of specifying what type of content is on a site and what it is about. Thats very useful for machines and we strongly advocate for the use of Schema.
However, does it work in LLMs. My original thought was yes! However it turns on thats not necessarily the case.
I turn again to Mark Williams-Cook who did a very interesting test, and this just goes to show why you need to look a little deeper before jumping to a conclusion.
Mark made up a fake website with a fake address which only lived in the schema – nowhere on the website did this address exist, except in the schema.
The LLMs returned information on this fake site and showed the address.
There you go – infallible proof that LLMs use Schema right?
Well no, not so fast.
When you consider what LLMs do and how the parse (read) content, they basically strip all the code away (and several words) and chunk it together (so there is no need to do this). So the schema was read, but it was just read as some text that existed. The LLM did not specifically chose and read schema (it didn’t even really know about the schema).
So you could reasonably conclude that LLMs do read schema, but it is not that straightforward.
As always the details matter, you need thorough testing and experimental validation of a theory to prove or disprove it.
Acronym Soup
Some of the AEO/GEO Acronym soup explained.
AEO – Answer Engine Optimization
GEO – Generative Engine Optimization
SEO – Search Engine Optimization – although I like the more modern take on it as Search EVERYWHERE Optimization.
SO – Search Optimization – Jake Ward / Cary Rose
AI – Artificial Intelligence
AIO – Artificial Intelligence Overview (a Google SERP feature)
LLM – Large Language Model
References & Resources
- June 22nd, 2006 – Google Tells All on AI Search – What the Documentation Actually Says – STOLBER Digital Marketing Agency, Trevor Stolber
- https://developers.google.com/search/docs/fundamentals/ai-optimization-guide
- September 26, 2025 – We Tested Query Fan-Out Optimization (Here’s What We Learned) – Semrush – Zach Paruch
- September 14, 2026 – Its not all Google – best AI Search Engines in 2026 – Zapier – Shubham Agarwal
- June 29, 2026 – Marketing AI Adoption Statistics (2026) – Omnibound – Sarah
- May 14, 2026 – Is Google losing their throne in the AI search race? – LinkedIn – Craig Rosenberg
- August 04, 2026 – How cats.txt Showed llms.txt Evidence is GEO Astrology – Mark Williams-Cook
- May 27, 2026 – Schema, LLMs and the Low Bar for “Evidence” in GEO – Mark Williams-Cook
